[prev in list] [next in list] [prev in thread] [next in thread] 

List:       sas-l
Subject:    Re: Exact binomial confidence limits
From:       vanbuski () ECLIPSE ! NET
Date:       2000-09-29 13:04:12
[Download RAW message or body]

This also works.

Mark Van Buskirk


** APPLY EXACT BINOMIAL 95% CONFIDENCE INTERVAL
**;

*  NOTE:  INCOMING DATASET HAS VARIABLES X,N;

DATA ALL;
  INPUT X N @@;
CARDS;
25 74 32 76
;

DATA ALL;
  SET ALL;
  ALPHA=0.05;
  C_LO=0.5*ALPHA;
  C_HI=1-C_LO;

  PCT=100*X/N;

  **.. 2-SIDED EXACT BINOMIAL 95% CONFIDENCE INTERVAL;
  D_HI=N-X;
  D_LO=D_HI+1;
  LO=100*(1/(1+D_LO/(X*FINV(C_LO,X+X,D_LO+D_LO))));
  HI=100*(1/(1+D_HI/((X+1)*FINV(C_HI,X+X+2,D_HI+D_HI))));

  **.. 1-SIDED C.I. IF PCT IS 0 OR 100;
  IF PCT=0 THEN
    DO;
      LO=0;
      HI=100*(1/(1+D_HI/((X+1)*FINV(1-ALPHA,X+X+2,D_HI+D_HI))));
      FLAG=1;
    END;
  IF PCT=100 THEN
    DO;
      LO=100*(1/(1+D_LO/(X*FINV(ALPHA,X+X,D_LO+D_LO))));
      HI=100;
      FLAG=1;
    END;

  DROP D_HI D_LO ALPHA C_LO C_HI FLAG;
  FORMAT LO HI 5.1;

PROC PRINT DATA=ALL; RUN;



Tim CHURCHES <TCHUR@DOH.HEALTH.NSW.GOV.AU> wrote in message
news:s9d0b23d.078@doh.health.nsw.gov.au...
> Dr John Whittington writes:
> > However, given that such an approach is at best an approximation, and
> will
> > only be a reasonable approximation over a certain limited range of
> values
> > of p, it would make much more sense to me to actually utilise an
> algorithm
> > which is calculates 'exact' CIs for any value of p.  This can be
> achieved
> > very easily in SAS, and if anyone is interested, I'll post the macro I
> > routinely use to do this.
>
> Herewith my take on this. I wrote this code a long time ago and looking at
it
> now makes me realise how SAS macros are no substitute for proper
> user-defined functions. Still, it does work.
>
> Tim Churches
> Sydney, Australia
>
> %macro BINCL(NUMVAR,DENVAR,CLEVEL,RATEVAR,LRATEVAR,URATEVAR) ;
>  %* Macro BINCL written by Tim Churches in an idle moment at New England
Region
>     Public Health Unit, Tamworth, New South Wales, Australia. 1 June,
1992. ;
>  %* BINCL calculates exact confidence limits for a proportion using the
>     inverse of the F distribution. See Armitage P and Berry G, Statistical
>     Methods in Medical Research,  2nd Ed. p119. Blackwell Scientific
Publications,
>     Oxford, 1987. ;
>  %* NUMVAR is the name of the variable which contains the numerator of the
>     proportion. ;
>  %* DENVAR is the name of the variable which contains the denominator of
the
>     proportion. ;
>  %* CLEVEL is the confidence level as a percentage ie to get 95% CIs,
>     use 95. ;
>  %* RATEVAR is the name of the variable to which you want the rate
(proportion)
>     to be assigned. ;
>  %* LRATEVAR and URATEVAR are the names of the variables to which you want
the
>     lower and upper limits of the rate (proportion) assigned. ;
>  drop onesidep ;
>  onesidep = (100 - &clevel)/200 ;
>  &ratevar =  &numvar / &denvar ;
>  &lratevar = &numvar /
>              (&numvar + (&denvar - &numvar +
1)*finv((1-onesidep),(2*&denvar - 2*&numvar + 2),(2*&numvar))) ;
>  &uratevar = (&numvar + 1) /
>              (&numvar + 1 + (&denvar -
&numvar)/finv((1-onesidep),(2*&numvar + 2),(2*&denvar - 2*&numvar))) ;
> %mend BINCL ;

[prev in list] [next in list] [prev in thread] [next in thread]